A DeadMod lyric video dropped three lines. The rest looked good. It still failed.
People know the words to a song. Missing one line makes the whole video feel broken.
The transcription provider returned full text, phrase segments, and word timestamps. They did not always agree. Punctuation moved. Contractions split. A segment ended early. Some words landed between phrases.
DeadMod now treats word timestamps as the timing record. Provider segments can suggest line breaks, but they do not decide what appears on screen. Each timed word joins the segment that covers its start time. If no segment covers it, it joins the nearest one. The renderer builds every displayed line from those timed words.
Automatic timing still misses things. Vocals, accents, mixes, and song structure will beat any general rule now and then. The editor lets someone fix the bad section instead of rebuilding the song, while preserving known timing where it can.
Preview and export use the same timing, fitting, and layout helpers. If the edit looks right in the editor, it should survive the exported video.